I'm not sure that we're seeing what you say we're seeing. On the M2, you can see clearly that the moment heavy processing begins on the performance cores, load DROPS on the efficiency cores. It is extremely unlikely that the efficiency cores are suddenly doing less work — I believe that what's actually happening is that the efficiency cores are being clocked up from their idle speed (around 700 MHz IIRC) to their full 2.4 GHz — to be prepared to handle any overspill from the performance cores, should they get overloaded.
AFAIK, this is how Apple Silicon deals with load balancing across CPU cores (for processes whose QoS settings allow assigning them to efficiency cores).
Does anyone know whether the CPU History window in Activity viewer shows current load per core as a percentage of power at current clockspeed or as a percentage of power at maximum clockspeed?
If the former, that would explain the M2 performance graphs. We cannot compare the M1, since those only show the CPU already under full load. That also doesn't explain why the M3 shows little difference, but that may just be due to the fact that there's so much E-power available in those six cores, that the efficiency tasks just never really register much, anyway. Or they've changed something in the thread handling in M3. I haven't found a detailed analysis yet on my usual source for this stuff, eclecticlight.co (highly recommended, btw.)
Because, again, they likely have completely different threading models.